On the shortest path and minimum spanning tree problems
Year: 2003
Type: dissertation
Abstract: The shortest path and minimum spanning tree problems are two of the classic \ntextbook problems in combinatorial optimization. They are simple to describe and \nadmit simple polynomial-time algorithms. However, despite years of concerted \nresearch effort, the asymptotic complexity of these problems remains unresolved. \nThe main contributions of this dissertation are a number of asymptotically \nfaster algorithms for the minimum spanning tree and shortest path problems. Of \nequal interest, we provide some clues as to why these problems are so diffcult. In \nparticular, we show why certain modern approaches to the problems are doomed \nto have super-linear complexity. A sampling of our results are listed below. We emphasize that all of our \nalgorithms work with general graphs, and make no restrictive assumptions on the \nnumerical representation of edge-lengths. \n A provably optimal deterministic minimum spanning tree algorithm. (We \ngive a constructive proof that the algorithmic complexity of the minimum \nspanning tree problem is equivalent to its decision-tree complexity.) \n An all-pairs shortest path algorithm for general graphs running in time \nO(mn + n2 \nlog log n), where m and n are the number of edges and vertices. \nThis provides the first improvement over approaches based on Dijkstra's \nalgorithm. \n An all-pairs shortest path algorithm for undirected graphs running in O(mn log α ) \ntime, where α = α (m, n) is the inverse-Ackermann function. \n A single-source shortest path algorithm running in O(m α +min{n log log r, n log n}) \ntime, where r bounds the ratio of any two edge lengths. For r polynomial \nin n this is O(m + n log log n), an improvement over Dijkstra's algorithm. \n An inverse-Ackermann style lower bound for the online minimum spanning \ntree veri cation problem. This is the rst inverse-Ackermann type lower \nbound for a comparison-based problem. \n An \nΩ (m + n log n) lower bound on any hierarchy-type single-source shortest \npath algorithm, implying that this type of algorithm cannot improve upon \nDijkstra's algorithm. (All of our shortest path algorithms are of the hierarchy \ntype.) \n The first parallel minimum spanning tree algorithm that is optimal w.r.t. to \nboth time and work. Our algorithm is for the EREW PRAM model. \n A parallel, expected linear-work minimum spanning tree algorithm using \nonly a polylogarithmic number of random bits. \n An O(mn log α ) bound on the comparison-addition complexity of all-pairs \nshortest paths. This is within a tiny log α factor of optimal when m = O(n).
Authors: Seth Pettie, Vijaya Ramachandran
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2160495255
